关键词: 玛咖, 红外光谱, 间隔偏最小二乘法, 遗传算法, 蛙跳算法

Abstract:

Based on Fourier transform infrared spectroscopy (FTIR), identification of the origin of 139 samples of maca
collected from Yunnan and Peru was conducted. The infrared spectra were preprocessed by multiple scattering correction
combined with second derivative and Norris smoothing. Through eliminating the noise spectral bands, the suitable number
of principal components was chose as eight. Based on the optimal number of principal components, by using interval partial
least squares (iPLS), the spectra in the range of 3 650.59–651.82 cm–1 was processed by optimization analysis. An iPLS-DA
classification model was built by screening the spectra of 98 samples in the ranges of 1 855.19–651.822, 3 054.69–2 756.78
and 3 650.59–3 353.6 cm–1. The R2, RMSEC and RMSEP of the model were 0.958 4, 0.785 8 and 1.164 2, respectively.
The verification with 41 samples indicated that the validation accuracy was consistent with that of the classification model
built using the original spectra, which was 87.80%. To further improve the accuracy of the classification model on the basis
of iPLS screening of spectral bands, the spectral information was optimized by genetic algorithm (GA) and shuffled frog
leaping algorithm (SFLA), respectively. The results showed that, through GA screening the frequency of spectral information
which was greater than 4 and 5, the filtered spectral data points were 62 and 29, respectively. Through SFLA screening the
probability of spectral information which was greater than 0.1 and 0.15, the filtered spectral data points were 77 and 27,
respectively. The validation results showed that the recognition efficiency of the classification model built by GA-PLS-DA
(62 data points) and GA-PLS-DA (29 data points) were 95.12% and 97.56%, respectively. The recognition efficiency of the
classification model built by SFLA-PLS-DA (77 data points) and SFLA-PLS-DA (27 data points) were 92.68% and 97.56%.
By comparing the above methods, we could find that the classification models built by iPLS-DA, GA-PLS-DA and SFLAPLS-
DA all had good prediction performance, of which the models built by GA-PLS-DA (29 data points) and SFLA-PLSDA
(27 data points) could more accurately identify the different origins of maca. The methods could provide a new way for
identification of the origin of maca with IR. The screening of the spectral variables could provide the basis for the difference
analysis of the chemical constitutes (components) in different origins of maca.

Key words: maca (Lepidium meyenii Walp.), infrared spectroscopy, interval partial least squares, genetic algorithm, shuffled frog leaping algorithm